agnes-the-ai-analyst/.github/workflows/ci.yml

name: CI

on:
  push:
    branches: [main, "feature/**"]
  pull_request:
    branches: [main]
  schedule:
    - cron: "0 3 * * *"  # Nightly at 03:00 UTC — runs docker-e2e
  workflow_dispatch:

jobs:
  test-shard:
    runs-on: ubuntu-latest
    strategy:
      fail-fast: false
      matrix:
        group: [1, 2, 3, 4]
    steps:
      - uses: actions/checkout@v6

      - uses: actions/setup-python@v6
        with:
          python-version: "3.13"

      - name: Install uv
        uses: astral-sh/setup-uv@v7

      - name: Install dependencies
        run: uv pip install --system ".[dev,server]"

      - name: Run tests (shard ${{ matrix.group }}/4)
        # pytest-split shards the suite across 4 parallel jobs, balanced by
        # the committed `.test_durations` file; `-n auto` parallelises
        # within each shard across the runner's cores. Regenerate durations
        # with `pytest tests/ --store-durations -n auto` when the suite
        # drifts enough that shards become uneven.
        run: pytest tests/ -v --tb=short -n auto --splits 4 --group ${{ matrix.group }}
        env:
          TESTING: "1"

  # Single required status check. Branch protection requires `test`, but the
  # matrix above publishes `test-shard (1..4)` — this job aggregates them
  # into one `test` result so no branch-protection change is needed.
  test:
    needs: test-shard
    if: always()
    runs-on: ubuntu-latest
    steps:
      - name: Verify all test shards passed
        run: |
          if [ "${{ needs.test-shard.result }}" != "success" ]; then
            echo "::error::test-shard result was '${{ needs.test-shard.result }}' — one or more shards failed"
            exit 1
          fi
          echo "All 4 test shards passed."          

  lint:
    # Advisory only (continue-on-error) — ruff + mypy surface issues but
    # never gate. Split out of release.yml's old test job; runs without the
    # full dependency install since neither tool needs it.
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v6

      - uses: actions/setup-python@v6
        with:
          python-version: "3.13"

      - name: Lint with ruff
        run: |
          pip install ruff
          ruff check . || true          
        continue-on-error: true

      - name: Type check with mypy
        run: |
          pip install mypy
          mypy src/ app/ cli/ connectors/ --ignore-missing-imports --no-error-summary || true          
        continue-on-error: true

  cli-wheel-clean-install:
    # Catches the "wheel METADATA conflicts with transitive deps under fresh
    # resolver" class — exactly what the workspace-only `[tool.uv]
    # override-dependencies` does NOT protect against. Builds the wheel the
    # way `release.yml` ships it to analysts (`uv build --wheel`), then
    # installs it into a fresh `python:3.13-slim` container with `uv tool
    # install` (the path the `/setup` page advertises) and asserts the
    # `agnes` binary actually launches. Without this, a regression like
    # 0.53.3's `kbcstorage>=0.9.0 → urllib3<2.0.0` cap silently caps the
    # wheel METADATA, every existing test passes (workspace overrides the
    # cap), and the break only surfaces on the next analyst's first install.
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v6

      - uses: actions/setup-python@v6
        with:
          python-version: "3.13"

      - name: Install uv
        uses: astral-sh/setup-uv@v7

      - name: Build wheel
        run: uv build --wheel --out-dir dist

      - name: Write CLI install assertion script
        run: |
          cat > /tmp/smoke.py <<'PY'
          import sys, urllib3
          try:
              import kbcstorage  # noqa: F401
              sys.exit("REGRESSION: kbcstorage leaked into the CLI wheel — should be in [server] extra only")
          except ImportError:
              pass
          maj, minor = (int(x) for x in urllib3.__version__.split(".")[:2])
          assert (maj, minor) >= (2, 7), f"urllib3 too old: {urllib3.__version__}"
          print(f"OK: kbcstorage absent, urllib3 {urllib3.__version__}")
          PY          

      - name: Smoke install in fresh python:3.13-slim
        run: |
          docker run --rm \
            -v "$PWD/dist:/wheels:ro" \
            -v /tmp/smoke.py:/smoke.py:ro \
            python:3.13-slim bash -c '
              set -euo pipefail
              apt-get update -qq && apt-get install -y -qq --no-install-recommends curl ca-certificates >/dev/null
              curl -LsSf https://astral.sh/uv/install.sh | sh > /dev/null 2>&1
              export PATH="$HOME/.local/bin:$PATH"
              WHEEL=$(ls /wheels/agnes_the_ai_analyst-*-py3-none-any.whl | head -1)
              uv tool install --force "$WHEEL"
              agnes --version
              agnes --help > /dev/null
              agnes catalog --help > /dev/null
              # Run the assertion in the same venv uv tool created
              "$HOME/.local/share/uv/tools/agnes-the-ai-analyst/bin/python" /smoke.py
            '          

  docker-build:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v6

      - name: Build Docker image
        run: docker build -t data-analyst:test .

  docker-e2e:
    runs-on: ubuntu-latest
    if: github.event_name == 'schedule' || github.event_name == 'workflow_dispatch'
    needs: docker-build
    steps:
      - uses: actions/checkout@v6

      - uses: actions/setup-python@v6
        with:
          python-version: "3.13"

      - name: Install uv
        uses: astral-sh/setup-uv@v7

      - name: Install dependencies
        run: uv pip install --system ".[dev,server]"

      - name: Start services
        run: |
          touch .env
          docker compose up -d --wait --wait-timeout 60          

      - name: Run Docker E2E tests
        run: pytest tests/ -v --tb=short -m docker --timeout=120
        env:
          TESTING: "1"

      - name: Stop services
        if: always()
        run: docker compose down